This article explores the limitations of Yann LeCun's approach to Large Language Models (LLMs) and the potential benefits of the Joint Embedding Architecture (JEPA). LeCun believes that existing LLM lacks a true understanding of the physical world and support for key intelligent capabilities such as persistent memory, reasoning, and planning. He emphasized the importance of building models that can deeply understand the world, and pointed out that JEPA's advantages in extracting abstract representations enable it to better learn the essential characteristics of the world, thus making up for the shortcomings of LLM.
Yann LeCun pointed out that although LLM has its uses, it cannot accurately understand the physical world and lacks support for basic intelligence features such as persistent memory, reasoning and planning. He discussed the possibility of building models with a deep understanding of the world, and introduced the advantages of Joint Embedding Architecture (JEPA) over LLM. JEPA can better extract abstract representations, enabling the system to essentially learn the abstract features of the world. .Taken together, LeCun's perspective highlights the future direction of the field of artificial intelligence, which is moving away from pure language processing and towards a deeper understanding of the physical world and abstract concepts. As a potential alternative, JEPA deserves further research and exploration in order to build more powerful and intelligent artificial intelligence systems.